What Happens When You Ask AI to Recommend Your Own Company?
A simple marketing experiment: ask ChatGPT, Claude, Perplexity and Google who they recommend in your category. Here's how to run the test properly and turn the results into an AI visibility audit.
Here's a simple marketing experiment.
Open ChatGPT.
Ask:
"What are the best companies for [your category]?"
Then ask:
"Which companies would you recommend for [your target audience]?"
Then:
"What are the best alternatives to [your biggest competitor]?"
Now look at the answers.
Is your company there?
If not, don't immediately assume something is wrong.
The interesting part is what happens next.
Run the test properly
Don't ask one question and call it an AI audit.
Create a list of questions your potential customers might actually ask.
For example, if you sell accounting software:
Category
"What are the best accounting software platforms?"
Audience
"What accounting software is best for UK startups?"
Use case
"What accounting software is best for a 20-person SaaS company?"
Comparison
"What are the best alternatives to X?"
Problem
"What accounting software makes it easiest to manage invoices and expenses?"
Now you've got five different ways someone might discover your company.
Test more than one AI assistant
Run the same questions across several platforms.
For example:
- ChatGPT
- Claude
- Perplexity
- Google AI search experiences
You may see different answers.
That's normal.
ChatGPT search uses web search and can provide cited sources. Perplexity says its search system researches the web and synthesizes information from multiple sources with citations.
Different systems can surface different information.
That's exactly why testing across platforms can be useful.
Record what happens
Create a simple spreadsheet.
| Prompt | ChatGPT | Claude | Perplexity | Google | | ------------------------ | ------- | ------ | ---------- | ------ | | Best accounting software | ✓ | ✗ | ✓ | ✓ | | Best for startups | ✗ | ✓ | ✗ | ✓ | | Alternatives to X | ✓ | ✗ | ✓ | ✗ |
Then add another column:
Was the information accurate?
Because appearance alone isn't enough.
Read the answer carefully
If your company appears, look at the description.
Does the AI understand:
- What you sell?
- Who you serve?
- Where you operate?
- Your main use cases?
- Your pricing?
- Your differentiators?
You might discover that the AI knows your company exists but has an outdated description.
That's a different problem from not appearing at all.
Then look at who gets mentioned instead
Your competitors can give you clues.
Not because you should copy them.
Because their visibility can show you what information exists in your category.
If competitors consistently appear when people ask:
"Best software for healthcare startups"
look at what the AI says about them.
Then investigate the sources.
Maybe they have:
- Healthcare case studies
- Industry-specific landing pages
- Reviews
- Press coverage
- Detailed product documentation
- Comparison articles
Now you've got something concrete to investigate.
Test your brand with difficult questions
Don't only ask easy questions.
Try questions where your positioning matters.
For example:
"Which [category] companies are best for businesses with fewer than 50 employees?"
"Which [category] companies specialize in healthcare?"
"What are good alternatives to [competitor] for a company that wants [specific feature]?"
These questions help reveal whether the AI understands the details that make your company relevant.
Test factual accuracy too
Ask:
"What does [company] do?"
Then compare the answer with your website.
Look for:
- Wrong product information
- Old pricing
- Wrong location
- Incorrect company size
- Old leadership information
- Confusion with another company
If the AI has the wrong information, investigate where that information may be coming from.
You may find an outdated directory, old article, or stale company profile that needs attention.
Don't expect a fixed answer
One of the biggest mistakes is treating AI visibility like a traditional ranking.
You might run the same prompt twice and receive different results.
That doesn't necessarily mean something changed on your website.
AI systems can use different searches, sources, context, and models.
OpenAI itself advises users to check cited sources because search results and citations can sometimes be incomplete, outdated, or incorrect.
That's why repeat testing is more useful than one screenshot.
Turn the experiment into an audit
Once you've run the test, categorize what you found.
Problem 1: We're invisible
Investigate discoverability, website accessibility, positioning, and available information.
Problem 2: We're mentioned but misunderstood
Improve your core messaging and product information.
Problem 3: We're mentioned but not recommended
Look at whether your site clearly explains use cases, audience, differentiation, and evidence.
Problem 4: Competitors dominate the answers
Investigate what information supports their visibility and where your own public information is weaker or less specific.
Problem 5: AI has outdated information
Find the source of the outdated information and update the places you control.
Run the experiment every month
The value comes from repetition.
Keep the same core questions.
Add new questions when your products, customers, competitors, or positioning change.
Track the answers.
Look for patterns.
You don't need to obsess over every individual response.
You're looking for broader changes in how AI systems understand and describe your business.
The bottom line
You don't need an expensive research project to get started with AI visibility.
You can learn a lot by asking the same questions your customers ask.
Who are the best companies in our category?
Who is best for our target customer?
What are the alternatives to our competitors?
What does AI think our company actually does?
The answers can reveal gaps in your website, positioning, content, and wider online presence.
And that's the useful part.
You're not testing AI for the sake of testing AI.
You're finding out how your business looks from the perspective of the systems increasingly being used to discover information.